01 · how you reach it
The engine is one file that runs on an ordinary laptop: no server, no licence counted per person, nothing that reaches out to the internet on its own. Every way in below runs that same file and gives the same answers.
how most teams use it the usual choice
You ask in plain words; the AI agent writes the model, runs it and reads the answer back, checking its own work against a real engine instead of guessing.
It connects over MCP, the open standard for giving an AI agent a new tool. Whether your agent speaks it is the one thing to ask your IT team first.
Available now
on its own
Point it at a model, get a results file back: the scheduled job at two in the morning when the order book has changed.
Available now
inside your own tools
Your analysts drive it from Python, hundreds of variants at once, into the tools they already use. Both write exactly the same model file.
Python available now · Java and Kotlin planned
02 · the six steps
Nobody on your side writes a simulation, maintains one, or is asked to prove one correct.
01
youa conversation to start
An export, a spreadsheet of routes and times, a maintenance log, or a conversation. Gaps become questions, not a reason to wait.
02
the AI agentminutes
A document your engineers can read, so the assumptions are visible instead of buried in code.
03
the AI agentminutes, until the gap closes
Run as your process actually ran, compared with what really happened: a gap is the finding. Ask for a stretch the model was never shown, because fitting one period is not prediction.
04
youone sentence
A second machine, a different order of work, one more shift: each is a small edit to the same document.
05
the engineovernight, even for a large plant
Each run many times over, so the verdict is better, worse, no meaningful difference, or not enough evidence yet.
06
the AI agentseconds
This one wins by about an hour a day; this one is inside the noise. The number and what it means, not a spreadsheet to interpret.
03 · the model
A page or two listing the machines, the routes and the times. Your engineers can read and correct it without learning a simulation language; the engine never changes, whatever you are modelling.
One question is one file, and a study is a folder of them. Keep the folder with your engineering records and you can show how a decision was reached, a year later.
{
"schema_version": 1,
"name": "paint-shop",
"horizon_seconds": 25920000.0,
"warmup_seconds": 2592000.0,
"resources": [
{
"id": "booth",
"servers": 1,
"changeovers": [
{ "from": "red", "to": "blue", "seconds": 2700.0 },
{ "from": "blue", "to": "red", "seconds": 2700.0 }
],
"entity_rule": { "ranks": [ { "weights": [["setup_cost", -1.0]] },
{ "weights": [["age", 1.0]] } ] }
}
],
"products": [
{ "id": "red_body",
"steps": [ { "resource": "booth", "setup": "red",
"process_seconds": { "kind": "constant", "value": 4200.0 } } ] },
{ "id": "blue_body",
"steps": [ { "resource": "booth", "setup": "blue",
"process_seconds": { "kind": "constant", "value": 4200.0 } } ] }
]
}